Team
Los sabrosos del código
Project Concept
PodAgent, an agent-based system designed to make YouTube and Spotify podcast content interactable through natural language
Entry
Status: Submitted
Last saved: May 09 at 4:54 PM -05
Team Roster
You must be registered for the event to view the team message board.
Juan Pablo Mejía Gómez Team Lead RSVP Approved
Full Stack Developer at ReshapeX
### Dev A — Backend MCP tools + Azure LLM wrapper
**Objective:** Implement MCP tools for LearnFlow using RAG-grounded structured output.
**Deliverables**
- MCP tools:
- `generate_plan(goal: str) -> Plan`
- `generate_section_content(section_id: str, goal: str) -> SectionContent`
- Keep/verify existing `answer_with_rag(question: str, context?: str) -> Answer`
- `get_episode(episode_id: str) -> Episode` (static lookup)
- Azure OpenAI client wrapper usable via dependency injection and stubbable in tests.
- Pydantic models that mirror the spec schemas.
**TDD checklist (pytest)**
1. **Contract tests (RED):**
- `generate_plan` returns a Pydantic `Plan` that serializes to the exact JSON shape.
- `generate_section_content` returns a Pydantic `SectionContent` with `flashcards[]` non-empty.
2. **LLM wrapper unit tests (RED):**
- Given a fake LLM response (string), parse + validate to model.
- Invalid JSON triggers retry/validation failure path.
3. **Implementation (GREEN):**
- Retrieval: embed goal/question → top-K chunks.
- Prompt: render from templates.
- Call Azure chat deployment.
- Parse + validate structured output.
**Subagent prompts**
- “Find existing MCP tool registration pattern + where to add new tools.”
- “Find existing OpenAI/embeddings code and how to swap to Azure OpenAI.”
**Acceptance**
- Running the server locally exposes the tools.
- Unit tests pass without Azure calls.
I am an Engineer in Systems and Informatics at the National University of Colombia, with a strong commitment to academic excellence and professional development. In addition to my studies, I have practical experience as a freelancer, collaborating on innovative software development projects.
My interests lie in areas such as Artificial Intelligence (including Machine Learning, Deep Learning, Agents, and Computer Vision) and Competitive Programming. I possess practical experience working with several frameworks such as Flask, FastAPI, Pytorch, Scikit-Learn, LangChain, LangGraph, and more. Additionally, I hold a certified C1 level in English
Artificial Intelligence, Machine Learning, Deep Learning, AI Agents, Computer Vision, Competitive Programming, RAG pipelines, Technical architecture, GTM, Partnerships, Product review, Design partners, Co-founders (Technical and Business), Founding Engineers, Investors
RAG AI-Powered Commercial Real Estate Lease Agreement Extractor with LangChain, AWS, …
Automated Defect Detection for IMUSA with SICK AI Camera and Computer Vision
Detecting Tuberculosis in X-Rays - Datacamp Challenge, with Transfer Learning, MLFlow, …
Medical AI Dashboard - XGBoost Literature Classification, with XGBoost, Scikit-Learn, ...
Multi-Agent Marketing Assistant with LangChain, LangGraph, FastAPI, Chainlit, …
Computer vision-based automated system as part of high school deg
Santiago Varela Vanegas RSVP Approved
Técnologo en sistemas at Universidad Nacional
### Dev B — Web routes + UI screens (exact spec)
**Objective:** Replace chat UX with LearnFlow flow across 5 routes.
**Routes to implement**
- `/` GoalInput
- `/plan` PlanView
- `/section/[id]` SectionView
- `/flashcards` FlashcardSession
- `/progress` ProgressDashboard
**Data flow**
- GoalInput “Generar mi plan” → call `generate_plan(goal)` → store in local state → navigate `/plan`.
- SectionView loads `generate_section_content(section_id, goal)` → renders accordions.
- “Iniciar estudio” → navigates to `/flashcards` seeded with flashcards.
**TDD checklist (vitest + RTL)**
1. Route smoke tests (RED): each route renders its required headline + primary CTA.
2. Navigation tests (RED): GoalInput → PlanView.
3. Component tests (RED):
- PlanView renders phases + DO/DON’T cards.
- SectionView renders Resumen/Glosario/Flashcards accordions.
- FlashcardSession shows progress counter + buttons.
4. Implementation (GREEN): build minimal components to satisfy tests.
**Subagent prompts**
- “Find existing Next.js app structure and any existing MCP client helper.”
- “Find Tailwind tokens / design primitives already present; avoid inventing new tokens.”
**Acceptance**
- All 5 routes render and navigate.
- No extra UX beyond the spec.
---
Santiago Varela is an Analista de automatización at SQA - Software Quality Assurance S.A, with 3 years of experience. He holds degrees in Ingeniería de sistemas e informática from Universidad Nacional de Colombia and Tecnólogo en análisis y desarrollo de sistemas from Institución universitaria Marco Fidel Suárez. Santiago focuses on BDD automation using the Screenplay pattern and SOLID principles, developing automated test suites with Java, Selenium, and Serenity BDD. He also manages code coverage with SonarQube, tracks defects in Jira, integrates tests into CI/CD pipelines, and mentors QA teams. He is looking for community and friendships.
AI engineer, machine learning, software development, quality assurance
Currently I’m not working in any particular proyect
Sebastián Gómez Zapata RSVP Approved
E-commerce Analyst Intern at Organización Corona
### Dev C — Ingestion/retrieval readiness + local progress + flashcards logic
**Objective:** Ensure local transcripts are ingested and retrievable; implement local-only progress store and flashcard session behavior.
**Deliverables**
- A one-command ingestion entrypoint for `podcasts/Huberman/`.
- Local progress store (React state or Zustand) matching `UserProgress` in spec.
- Flashcard session logic:
- flip card
- next card
- record known/unknown results
- progress bar state
**TDD checklist**
1. Store tests (RED): actions update immutable state correctly.
2. Flashcard reducer/unit tests (RED):
- marking known/unknown records result
- advancing increments index
3. Optional integration test: ingest 1 fixture transcript → retrieval returns at least 1 chunk.
**Subagent prompts**
- “Find ingestion entrypoints and current vector store configuration for local dev.”
- “Find existing state management patterns in web app; match conventions.”
**Acceptance**
- Ingestion can be run locally (documented command).
- Flashcard session updates progress locally (no backend persistence).
---
I’m Sebastián, a systems engineer from the National University of Colombia with a strong interest in artificial intelligence, data, and building practical solutions. I have experience in programming, data structures, and financial analysis, and I like approaching problems in a structured and logical way.
Right now, I’m focused on learning and applying AI concepts—especially large language models, agents, and automation—while building real-world projects.
I value long-term thinking, continuous learning, and hands-on practice. I complement my learning through platforms like Platzi and DataCamp, always aiming to turn what I learn into something real
IA, machine learning, LLMS, Agents, Financial Analysis
I’m currently working on an AI agent project inspired by the podcast of Andrew Huberman. His content is incredibly rigorous and scientifically grounded, and it has personally been very useful for improving habits related to sleep, focus, and overall well-being.
However, the challenge is that the amount of information is massive—hundreds of long-form episodes, each packed with dense insights. It’s difficult to efficiently extract and apply that knowledge in a practical way.
To address this, I’m